Large Margin LAD Models and LAD-based Regression

Tiberius Bonates



We discuss some recent developments in Logical Analysis of

Data (LAD), with a focus on (i) the use of the concept of

large margin classifiers to build parameter-free LAD

classification models, and (ii) the extension of the LAD

methodology to address regression problems.

Both ideas rely on simple optimization models, strongly based

on the original formulation proposed by Hammer et al. in the

seminal LAD papers.

We compare the proposed techniques to standard classification

and regression techniques through a series of experiments on

datasets from the UCI repository. Finally, we discuss some

extensions of this work and of the LAD methodology

in general.